Dose reconstruction supports the interpretation of decreased abundance of mammals in the Chernobyl Exclusion Zone

The principle of dose reconstruction supposes to gather a significant amount of data in multiple areas, from radiological measurements to ecological information for each species. Since our primary dataset was not acquired with this objective in mind, we faced a lack of information for some descriptors. We filled these missing values using reasonable assumptions founded on scientific justification as described below.

Study sites and mammals tracks

We re-analyzed the dataset described by Møller and Mousseau¹. They used foot prints following fresh snow fall to estimate abundances of mammals, as counted by a single observer on a single ca. 2 day period of 3–4 February 2009. A total of 161 line transects were surveyed, each with a length of 100 m. Transects were separated by at least 50 m (but usually 100–500 m), and were located along roadsides (Fig. 1). A rigorous examination of the consistency and homogeneity of the dataset led us to exclude 16 transects from our analysis. Sixteen of these transects (see SI) were investigated during a different period (January 21 or February 17 and 18, 2009), applying a different experimental logic. They did not belong to the same sampling plan, and their use would have reduced the statistical significance of our analysis. This reduction did not affect the diversity of mammal species observed. The revised dataset included the abundance of 12 species of mammals distributed over 145 transects (Fig. 7). Foxes were the most frequently observed species, followed by wolves. Large prey (deer, horse, moose, and wild boar) had count numbers from about ten to twenty individuals.

About 18% of the transects were devoid of mammal activity during the study period (Fig. 8). Predators tended to be more widely distributed than prey, being observed in 72% of transects while prey were not seen in about half the transects. Most transects (> 80%) had fewer than 4 individuals. The paucity of observations prevented a more refined analysis of the taxonomic structure of these data.

We used a number of additional descriptors for each site as covariates in our analysis, as described by Møller and Mousseau¹. These included temporal descriptors (i.e. time of day the counting started, time since the last snow fall) and environmental factors (percentage cover with grass, bushes and trees to the nearest 5%). Ambient radiation levels were determined from averaged repeated measurements (2–3) at ground level with a hand-held Geiger counter (Model: Inspector, SE International, Inc., Summertown, TN, USA).

Dose reconstruction

The principles for reconstruction of total radiological doses absorbed by animals were described in detail in a previous similar study of dose reconstruction for birds in the Fukushima area⁵ (SI). According to this method, we estimated for the mammal species j and the radionuclide r the external and internal dose rates (respectively EDR(j,r) and IDR(j,r); µGy h⁻¹, see supporting Excel® file) absorbed by mammals of a given species. We used radionuclide activity concentrations measured in soils and calculated for animals, multiplied by the ad-hoc Dose Coefficients (DCs, Table S8). These coefficients, specific for the combination of radionuclides, species and irradiation pathway, were determined per radionuclide and daughter element(s) for adult stage of each species with the EDEN v3.1 software²⁷.

The internal and external irradiation dose rates absorbed by mammal species were calculated according to the equations described elsewhere⁵ (and recalled in SI). The total dose rate absorbed by the mammal species j, TDR_j, is the sum of the internal and external dose rates assessed for each radionuclide, applied to all radionuclides of interest (SI Excel® file).

We assessed the total dose to a given mammal species j considering that adult individuals are exposed to ionizing radiation at a constant dose rate during a period that corresponds to the generation time L_Gj of the species (average age of parents of the current cohort reflecting the turnover rate of breeding individuals in a population; SI). Consequently, the total dose absorbed by each species j at a given transect i noted td_j,i resulted from

This total dose was calculated for each species and each transect, whether the presence of the species on this given transect has been confirmed or not. This is the theoretical total dose the animal would receive if living there.

The final objective of the re-analysis was to study how abundance of different groups of mammals (all mammals, prey and predators) identified at the study sites vary according to their exposure to ionizing radiation. We needed a unique value of dose per transect, representative of the average exposure of all animals indirectly observed along this transect. The td_j,i values presented large ranges of variation (intra-transect ratio from ca. 30—transect n°10—to 6,000—transects n°83, 84 and 92). Their geometric mean for all species on a given transect i was calculated as the most relevant indicator, named Transect Total Dose and abbreviated TTD_i. Calculating the mean on all species, whatever they have been counted on the transect or not, gives a highly representative estimate of the level of exposure on the considered transect, not of the exposure of the counted species. This was also justified because we considered the zero count as relevant information. Such a number may have two origins: either the species has never occupied the surroundings of the transect (e.g. area not suitable for its needs) or it has disappeared. In both cases, the series of possible confounding variables considered in our statistical analyses will allow us to include this information (see “Methods”, Statistics).We used the geometric mean value in order to limit the influence of extreme values on the results²⁸.

Mammal species and associated assumptions

We deliberately chose to limit our analysis to adults in order to minimize the assumptions required to achieve our calculations. It is generally recognized that juveniles may be more sensitive to exposure to pollutants than adults. Juvenile development and growth mobilize resources that are no longer available for their protection. Juveniles differ from adults in their diet, behavior and physiological characteristics. Moreover, these characteristics change with time from birth to maturity. Such changes can have large implications in terms of dose reconstruction and associated uncertainties. Thus it is necessary to identify periods of development during which individual characteristics can be considered constant, and to be able to collect data corresponding to the needs for dose reconstruction. This approach is possible for a single species, but would be much more speculative for all 12 of our species of interest. Since our understanding of adult life history is likely to be more robust than that of juveniles for the purposes of dose reconstruction, we have ignored juvenile stages for this analysis. Moreover, time from birth to maturity is generally short with regard to generation time (Table S1), and discounting the corresponding contribution to the total dose would underestimate its actual value in a way that makes our results an acceptable proxy for the quantification of the response of mammals to their exposure to ionizing radiation.

For each of the 12 species under consideration, DC calculation required us to simplify the representation of adults as ellipsoids of known mass and size (geometric characteristics, Tables S1 and S2), and to define media elementary composition (Table S3). In the same way, a basic animal life style was described considering the time spent (i) in a burrow if relevant for the species and (ii) standing or lying on soil for all species (Table S4). Finally, as much attention as possible was paid to the species’ diet (omnivorous, carnivorous or herbivorous) to select the most appropriate value for the concentration ratios (CR) required to quantify the radionuclide aggregated transfer from soil to the animal (Table S5). When available, site-specific CRs were preferentially used, to reduce the large uncertainty associated with the choice of a CR value. This uncertainty is a well-known weakness of the assessment of activity concentrations in animals applying the equilibrium approach^29,30. By default for site-specific data, the choice was made to refer to best-estimates published in an international compilation of data³¹. All data depending on the nature of the radionuclide were collected or calculated for the elements Cs and Sr and their isotopes present in the accidental releases for the Chernobyl NPP accident (Table S6). Since dose (rate) is additive in terms of the resulting effects of exposure to ionizing radiation, it is essential to exhaustively characterize the source of radioactivity under examination in terms of quality and quantity.

Soil contamination data

We conducted two preliminary studies to streamline and optimize data collection, and to limit the assumptions required to fill potential data gaps. First, we explored the depth of contaminated soil for consideration in the calculation of DCs. A potential maximal depth of 10 cm has been reported for the radioisotopes characteristic of the accident fallout (Table S6), which is in agreement with observed and predicted contamination profiles for ¹³⁷Cs (Fig. S1). In the end we used a 20 cm layer, in a conservative but realistic way, as increasing reasonably the soil depth increases the amount of radioactivity to which mammals could be exposed. Despite the much larger original spectrum of radionuclides, it is largely assumed that today both ¹³⁷Cs and ⁹⁰Sr should be the main markers of the impact on the environment of the NPP accident, due to the emitted quantities and their radioactive half-lives. Radionuclides contribute very differently to the total dose absorbed by animals depending on the energies and nature of their emissions³². We thus secondly investigated the role of the 10 radioisotopes for which we found activity concentrations in soil considering their realistic extreme values in a given location (Table S7). We assessed the corresponding total dose rates on one hand for the pair ¹³⁷Cs + ⁹⁰Sr and their daughters, and on the other hand for the remaining radionuclides, for two contrasting mammal species, a small carnivore and a large herbivore (Fig. S2). Whatever the scenario, the dose rate due to ¹³⁷Cs and ⁹⁰Sr represents at least 94% of the total exposure. We assumed that other radionuclides can be ignored without significantly skewing our results, taking into account all the associated uncertainties. This considerably limits the data search and collection (focused on selected isotopes, e.g. DC values, Table S8) as well as the assumptions necessary to achieve the dose reconstruction. It results in a reduction of calculations needed but also of conservatism of the approach, while keeping it at a level sufficient for our needs.

Measurements of soil radionuclide activity concentrations have been extensively conducted in the CEZ and around since the accident. To best cover the spatial and temporal scales of our study, we combined different data sources^6,33,34,35 (plus the REDAC database, V. Kashparov, personal communication). ¹³⁷Cs and ⁹⁰Sr soil activities were assessed for each transect. We took into account both the transect length (100 m) and the species home range (Table S1) to define a potential exposure area for each species present on a given transect (dosimetry area, Table S2). This circular area is centered on the transect origin, located by its GPS coordinates, with a radius of 100 m (transect length) plus the radius of the species’ home range (Fig. 9). Using GIS, we crossed referenced this information with the geo-located contamination data from all the references identified. When several measurements were available for the same dosimetry area, we retained their extreme and mean values (i.e. in general three different values per dosimetry area). When only one single measurement was available, we used these data for both extreme and mean values. When no data were available in a given area, soil activity was assumed equal to the one measured at the nearest soil sampling point. The radioactive decay occurring during the period of dose reconstruction (i.e. the generation time) was ignored with regard to the ratio between the generation times (highest value for the red deer L_G: 5,210 days, ca. 14 years) and the radionuclide periods (about 30 years for both ¹³⁷Cs and ⁹⁰Sr). This assumption contributes to the conservatism of the approach. The final dataset included three values of soil activity concentration per radionuclide (¹³⁷Cs and ⁹⁰Sr) for each species on each transect (i.e. more than 10,000 values). We arbitrarily decided not to use more complex data treatment such as krieging. Due to the highly heterogenous “leopard skin” pattern of the soil contamination, we considered such approaches not particularly robust as they give an apparent continuity to soil contamination between measured values. Using only actually measured values helped to limit the number of assumptions required by our calculation, already high. We acknowledge however that a spatialized statistical approach to better assess the soil contamination is an interesting perspective to refine the dose reconstruction.

Uncertainties

The first source of uncertainty in this study was its field protocol, which did not allow screening of the possibility of a double count of the same animal. This is a well-known weakness of such census methods. This “old fashioned” approach of field counting was largely applied in the past, as it is something relatively easy and simple to implement, requiring relatively few resources in contrast to more technological methods. The related and inherent disadvantage is the uncertainty around the count that is difficult to quantify. The way the census was conducted was though to reduce this uncertainty, by exploring somewhat distant transects in a short period of time. This is not a guarantee that individuals have not been counted more than once, but the application of recommended best practices when using such methods.

Other sources of uncertainty appear in such a dose reconstruction, which is a highly uncertain exercise requiring numerous assumptions. The use of CRs was previously acknowledged as a major source of uncertainty that we managed by constraining the value by the diet and using preferentially site-specific values or by default best-estimates such as CR values provided by the IAEA³¹. Using these values under-predicted the transfer of ¹³⁷Cs to predatory versus to prey species³¹. Wood et al.³⁶ reported from previous studies that the transfer of cesium to carnivorous species such as those classified in our study as predatory was suggested to be higher than for mammals at lower trophic levels. At the opposite, values summarized by the IAEA showed a significantly higher transfer of cesium to omnivorous and herbivorous mammals, as data in this database relate³¹ mainly to insectivorous small mammals rather than to species similar to our predatory mammals. In the absence of site-specific CR values, the use of best-estimates remained the best option.

We applied this logic to any other ecological or biological parameter required by the dose calculation (home range, animal size, etc.). Our choice was to make an assessment as specific and realistic as possible, without propagating uncertainty characterized at a global scale. For all parameters, there were insufficient local data to characterize their local variation. Soil activities showed rapid spatial change. This is a well-known characteristic of the contamination in the CEZ and it was the only data that we were able to characterize locally in terms of range of variation. We decided to consider only this site-specific uncertainty in our calculations.

A last source of uncertainty in approaches such as the one applied here is the existence of confounding factors. A number of additional variables are known as potentially affecting mammal abundance, such as environmental characteristics or human activities. The first problem consists in identifying these variables, and then to characterize them. What are the necessary and sufficient parameters to collect, when and how? Regarding the description of the environment, the minimal dataset usually acknowledged as relevant has been collected during the census (soil cover, time of observation, time elapsed since the last snowfall). If time data provide objective information, soil cover is observer dependent. This uncertainty was reduced due to observations done by a single observer. This ensured a high comparability between transects. The interaction between animal abundance and human activities is somewhat more complex to characterize. The nature of the CEZ led us to consider only the potential for repulsion of industrial activity linked with the NPP or attraction of farming areas. Characterizing the latest areas was highly uncertain (see dedicated paragraph in SI). There may be a significant time shift between the time of required data were acquired and the time of census. There may be also problems of spatial definition due to labels used in the available sources of information.

Statistics

The re-analysis of the dataset gave the opportunity to investigate the role of complementary data related to the impact of human activities. Potential spatial interactions between industrial and farming activities, present in the CEZ, and exposure areas of all or parts of the mammals were considered before to be dismissed as non-significant in the conditions of this study (see SI). The set of confounding variables finally retained was the same as for the initial study, that is to say the environmental descriptors that were recorded during the census (time of observation, time since snowfall and soil cover type expressed as % of tree, bush and grass).

All statistical analyses were performed in R³⁷. We first tested the variation in mammal abundance with the TTD_i increase through the development of a Generalized Linear Mixed Model (GLMM), assuming Poisson error distribution. The main predictor (TTD_i) was log-transformed and then centered on the mean and scaled by the standard deviation. The multicollinearity between possible confounding variables was checked through the Pearson correlation coefficient (omitted if Pearson correlation coefficient > 0.85, and using³⁸ a Variance Inflation Factor < 3), in order to include in the non collinear standardized variables in the model. Site was introduced as a random effect, and transects were treated as statistically independent observations. Secondly, the variation in abundance of mammal subgroups (predators and prey) was tested in the same way with regard to the TTD_i increase. The same fixed and random effects were considered. The last analysis aimed to analyze the variation in predator abundance with TTD_i, while considering the number of prey as an additional fixed effect.

Finally, four GLMMs were used to test the following hypotheses (where No means number):

All GLMMs were fitted with the lme4 package³⁹. For each model, we computed adjusted p-values using the multcomp package in order to take into account the multiple null hypotheses tested simultaneously⁴⁰.

Source: Ecology - nature.com